The Research For Double Compression And Resampling Detection Algorithm Of Digital Audio

With the rapid development of the Internet, the widely use of smart phones, digital audios has been used by everyone. While some convenient softwares about digital audio processing which have simple interfaces can be seen everywhere on the internet, making people tamper digital audios easly and leave little perceptible traces. Identify the authentication and integrity of audios has become a hot research topic.Many experts is working for decting audio editing operations such as resampling, splice,double compression, format conversion, to judge the authentication and integrity of digital audios.The main content of this article is aim to solve certain existed problems about audio authenticity. We have proposed two effective methods to detect double AMR compression and audio resampling. Thesis of this paper is as following:(1) The paper gives a comprehensive summary of the development of nearly30years of digital audio forensics,mainly related about research background and status. Given some most used digital audio fomats a brief description, and described existing audio editor softwares and a variety of tamper methods. At last,present common methods to detect audio forgeries.(2) Through analysising compression theory of AMR format audio, we proposed a algorithm based on support vector machine (SVM) for AMR compression detection. This algorithm analyzes the change of frequency energy distribution and the frequency composition after double AMR compression.It can be found that the energy distribution of AMR files through double compressed is difference from a single compressed AMR files,so statistical features reflecting this can be extracted from AMR audio.Experimental results show that our method can effectively identify the AMR file whether through double compression.(3) This paper analyzed the resampled audio signal.proposed a detection method based on variance of the second derivative of the signal. We can get the peak of the variance of the second derivative of audio signal, determine whether the audio has been interpolated,then calculate the interpolation factor. Experimental results show that our method can effectivly detect local and global interpolation signals with no time-consuming and other limitations.At last, we summarized the main contents about this paper and given the possible research areas of the audio forensic in the future.